md: reject a re-add request that cannot be honoured.
[deliverable/linux.git] / drivers / md / bitmap.c
CommitLineData
32a7627c
N
1/*
2 * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003
3 *
4 * bitmap_create - sets up the bitmap structure
5 * bitmap_destroy - destroys the bitmap structure
6 *
7 * additions, Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.:
8 * - added disk storage for bitmap
9 * - changes to allow various bitmap chunk sizes
32a7627c
N
10 */
11
12/*
13 * Still to do:
14 *
15 * flush after percent set rather than just time based. (maybe both).
32a7627c
N
16 */
17
bff61975 18#include <linux/blkdev.h>
32a7627c 19#include <linux/module.h>
32a7627c
N
20#include <linux/errno.h>
21#include <linux/slab.h>
22#include <linux/init.h>
32a7627c
N
23#include <linux/timer.h>
24#include <linux/sched.h>
25#include <linux/list.h>
26#include <linux/file.h>
27#include <linux/mount.h>
28#include <linux/buffer_head.h>
43b2e5d8 29#include "md.h"
ef740c37 30#include "bitmap.h"
32a7627c 31
e384e585 32#include <linux/dm-dirty-log.h>
32a7627c
N
33/* debug macros */
34
35#define DEBUG 0
36
37#if DEBUG
38/* these are for debugging purposes only! */
39
40/* define one and only one of these */
41#define INJECT_FAULTS_1 0 /* cause bitmap_alloc_page to fail always */
42#define INJECT_FAULTS_2 0 /* cause bitmap file to be kicked when first bit set*/
43#define INJECT_FAULTS_3 0 /* treat bitmap file as kicked at init time */
44#define INJECT_FAULTS_4 0 /* undef */
45#define INJECT_FAULTS_5 0 /* undef */
46#define INJECT_FAULTS_6 0
47
48/* if these are defined, the driver will fail! debug only */
49#define INJECT_FATAL_FAULT_1 0 /* fail kmalloc, causing bitmap_create to fail */
50#define INJECT_FATAL_FAULT_2 0 /* undef */
51#define INJECT_FATAL_FAULT_3 0 /* undef */
52#endif
53
32a7627c
N
54#ifndef PRINTK
55# if DEBUG > 0
56# define PRINTK(x...) printk(KERN_DEBUG x)
57# else
58# define PRINTK(x...)
59# endif
60#endif
61
ac2f40be 62static inline char *bmname(struct bitmap *bitmap)
32a7627c
N
63{
64 return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
65}
66
32a7627c
N
67/*
68 * just a placeholder - calls kmalloc for bitmap pages
69 */
70static unsigned char *bitmap_alloc_page(struct bitmap *bitmap)
71{
72 unsigned char *page;
73
44456d37 74#ifdef INJECT_FAULTS_1
32a7627c
N
75 page = NULL;
76#else
ac2f40be 77 page = kzalloc(PAGE_SIZE, GFP_NOIO);
32a7627c
N
78#endif
79 if (!page)
80 printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap));
81 else
a654b9d8 82 PRINTK("%s: bitmap_alloc_page: allocated page at %p\n",
32a7627c
N
83 bmname(bitmap), page);
84 return page;
85}
86
87/*
88 * for now just a placeholder -- just calls kfree for bitmap pages
89 */
90static void bitmap_free_page(struct bitmap *bitmap, unsigned char *page)
91{
92 PRINTK("%s: bitmap_free_page: free page %p\n", bmname(bitmap), page);
93 kfree(page);
94}
95
96/*
97 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
98 *
99 * 1) check to see if this page is allocated, if it's not then try to alloc
100 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
101 * page pointer directly as a counter
102 *
103 * if we find our page, we increment the page's refcount so that it stays
104 * allocated while we're using it
105 */
ac2f40be
N
106static int bitmap_checkpage(struct bitmap *bitmap,
107 unsigned long page, int create)
ee305ace
N
108__releases(bitmap->lock)
109__acquires(bitmap->lock)
32a7627c
N
110{
111 unsigned char *mappage;
112
113 if (page >= bitmap->pages) {
1187cf0a
N
114 /* This can happen if bitmap_start_sync goes beyond
115 * End-of-device while looking for a whole page.
116 * It is harmless.
117 */
32a7627c
N
118 return -EINVAL;
119 }
120
32a7627c
N
121 if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
122 return 0;
123
124 if (bitmap->bp[page].map) /* page is already allocated, just return */
125 return 0;
126
127 if (!create)
128 return -ENOENT;
129
32a7627c
N
130 /* this page has not been allocated yet */
131
ac2f40be
N
132 spin_unlock_irq(&bitmap->lock);
133 mappage = bitmap_alloc_page(bitmap);
134 spin_lock_irq(&bitmap->lock);
135
136 if (mappage == NULL) {
32a7627c
N
137 PRINTK("%s: bitmap map page allocation failed, hijacking\n",
138 bmname(bitmap));
139 /* failed - set the hijacked flag so that we can use the
140 * pointer as a counter */
32a7627c
N
141 if (!bitmap->bp[page].map)
142 bitmap->bp[page].hijacked = 1;
ac2f40be
N
143 } else if (bitmap->bp[page].map ||
144 bitmap->bp[page].hijacked) {
32a7627c
N
145 /* somebody beat us to getting the page */
146 bitmap_free_page(bitmap, mappage);
147 return 0;
ac2f40be 148 } else {
32a7627c 149
ac2f40be 150 /* no page was in place and we have one, so install it */
32a7627c 151
ac2f40be
N
152 bitmap->bp[page].map = mappage;
153 bitmap->missing_pages--;
154 }
32a7627c
N
155 return 0;
156}
157
32a7627c
N
158/* if page is completely empty, put it back on the free list, or dealloc it */
159/* if page was hijacked, unmark the flag so it might get alloced next time */
160/* Note: lock should be held when calling this */
858119e1 161static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page)
32a7627c
N
162{
163 char *ptr;
164
165 if (bitmap->bp[page].count) /* page is still busy */
166 return;
167
168 /* page is no longer in use, it can be released */
169
170 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
171 bitmap->bp[page].hijacked = 0;
172 bitmap->bp[page].map = NULL;
ac2f40be
N
173 } else {
174 /* normal case, free the page */
175 ptr = bitmap->bp[page].map;
176 bitmap->bp[page].map = NULL;
177 bitmap->missing_pages++;
178 bitmap_free_page(bitmap, ptr);
32a7627c 179 }
32a7627c
N
180}
181
32a7627c
N
182/*
183 * bitmap file handling - read and write the bitmap file and its superblock
184 */
185
32a7627c
N
186/*
187 * basic page I/O operations
188 */
189
a654b9d8 190/* IO operations when bitmap is stored near all superblocks */
f6af949c 191static struct page *read_sb_page(mddev_t *mddev, loff_t offset,
a2ed9615
N
192 struct page *page,
193 unsigned long index, int size)
a654b9d8
N
194{
195 /* choose a good rdev and read the page from there */
196
197 mdk_rdev_t *rdev;
a654b9d8 198 sector_t target;
ac2f40be 199 int did_alloc = 0;
a654b9d8 200
ac2f40be 201 if (!page) {
a2ed9615 202 page = alloc_page(GFP_KERNEL);
ac2f40be
N
203 if (!page)
204 return ERR_PTR(-ENOMEM);
205 did_alloc = 1;
206 }
a654b9d8 207
159ec1fc 208 list_for_each_entry(rdev, &mddev->disks, same_set) {
b2d444d7
N
209 if (! test_bit(In_sync, &rdev->flags)
210 || test_bit(Faulty, &rdev->flags))
ab904d63
N
211 continue;
212
ccebd4c4 213 target = offset + index * (PAGE_SIZE/512);
a654b9d8 214
2b193363 215 if (sync_page_io(rdev, target,
e1defc4f 216 roundup(size, bdev_logical_block_size(rdev->bdev)),
ccebd4c4 217 page, READ, true)) {
ab904d63 218 page->index = index;
ce25c31b
N
219 attach_page_buffers(page, NULL); /* so that free_buffer will
220 * quietly no-op */
ab904d63
N
221 return page;
222 }
223 }
ac2f40be
N
224 if (did_alloc)
225 put_page(page);
ab904d63 226 return ERR_PTR(-EIO);
a654b9d8 227
a654b9d8
N
228}
229
b2d2c4ce
N
230static mdk_rdev_t *next_active_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
231{
232 /* Iterate the disks of an mddev, using rcu to protect access to the
233 * linked list, and raising the refcount of devices we return to ensure
234 * they don't disappear while in use.
235 * As devices are only added or removed when raid_disk is < 0 and
236 * nr_pending is 0 and In_sync is clear, the entries we return will
237 * still be in the same position on the list when we re-enter
238 * list_for_each_continue_rcu.
239 */
240 struct list_head *pos;
241 rcu_read_lock();
242 if (rdev == NULL)
243 /* start at the beginning */
244 pos = &mddev->disks;
245 else {
246 /* release the previous rdev and start from there. */
247 rdev_dec_pending(rdev, mddev);
248 pos = &rdev->same_set;
249 }
250 list_for_each_continue_rcu(pos, &mddev->disks) {
251 rdev = list_entry(pos, mdk_rdev_t, same_set);
252 if (rdev->raid_disk >= 0 &&
b2d2c4ce
N
253 !test_bit(Faulty, &rdev->flags)) {
254 /* this is a usable devices */
255 atomic_inc(&rdev->nr_pending);
256 rcu_read_unlock();
257 return rdev;
258 }
259 }
260 rcu_read_unlock();
261 return NULL;
262}
263
ab6085c7 264static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
a654b9d8 265{
b2d2c4ce 266 mdk_rdev_t *rdev = NULL;
a6ff7e08 267 struct block_device *bdev;
ab6085c7 268 mddev_t *mddev = bitmap->mddev;
a654b9d8 269
b2d2c4ce 270 while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
ac2f40be
N
271 int size = PAGE_SIZE;
272 loff_t offset = mddev->bitmap_info.offset;
a6ff7e08
JB
273
274 bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
275
ac2f40be
N
276 if (page->index == bitmap->file_pages-1)
277 size = roundup(bitmap->last_page_size,
a6ff7e08 278 bdev_logical_block_size(bdev));
ac2f40be
N
279 /* Just make sure we aren't corrupting data or
280 * metadata
281 */
282 if (mddev->external) {
283 /* Bitmap could be anywhere. */
284 if (rdev->sb_start + offset + (page->index
285 * (PAGE_SIZE/512))
286 > rdev->data_offset
287 &&
288 rdev->sb_start + offset
289 < (rdev->data_offset + mddev->dev_sectors
290 + (PAGE_SIZE/512)))
291 goto bad_alignment;
292 } else if (offset < 0) {
293 /* DATA BITMAP METADATA */
294 if (offset
295 + (long)(page->index * (PAGE_SIZE/512))
296 + size/512 > 0)
297 /* bitmap runs in to metadata */
298 goto bad_alignment;
299 if (rdev->data_offset + mddev->dev_sectors
300 > rdev->sb_start + offset)
301 /* data runs in to bitmap */
302 goto bad_alignment;
303 } else if (rdev->sb_start < rdev->data_offset) {
304 /* METADATA BITMAP DATA */
305 if (rdev->sb_start
306 + offset
307 + page->index*(PAGE_SIZE/512) + size/512
308 > rdev->data_offset)
309 /* bitmap runs in to data */
310 goto bad_alignment;
311 } else {
312 /* DATA METADATA BITMAP - no problems */
313 }
314 md_super_write(mddev, rdev,
315 rdev->sb_start + offset
316 + page->index * (PAGE_SIZE/512),
317 size,
318 page);
b2d2c4ce 319 }
a654b9d8
N
320
321 if (wait)
a9701a30 322 md_super_wait(mddev);
a654b9d8 323 return 0;
4b80991c
N
324
325 bad_alignment:
4b80991c 326 return -EINVAL;
a654b9d8
N
327}
328
4ad13663 329static void bitmap_file_kick(struct bitmap *bitmap);
32a7627c 330/*
a654b9d8 331 * write out a page to a file
32a7627c 332 */
4ad13663 333static void write_page(struct bitmap *bitmap, struct page *page, int wait)
32a7627c 334{
d785a06a 335 struct buffer_head *bh;
32a7627c 336
f0d76d70
N
337 if (bitmap->file == NULL) {
338 switch (write_sb_page(bitmap, page, wait)) {
339 case -EINVAL:
340 bitmap->flags |= BITMAP_WRITE_ERROR;
f0d76d70 341 }
4ad13663 342 } else {
a654b9d8 343
4ad13663 344 bh = page_buffers(page);
c708443c 345
4ad13663
N
346 while (bh && bh->b_blocknr) {
347 atomic_inc(&bitmap->pending_writes);
348 set_buffer_locked(bh);
349 set_buffer_mapped(bh);
721a9602 350 submit_bh(WRITE | REQ_SYNC, bh);
4ad13663
N
351 bh = bh->b_this_page;
352 }
d785a06a 353
ac2f40be 354 if (wait)
4ad13663
N
355 wait_event(bitmap->write_wait,
356 atomic_read(&bitmap->pending_writes)==0);
8a5e9cf1 357 }
4ad13663
N
358 if (bitmap->flags & BITMAP_WRITE_ERROR)
359 bitmap_file_kick(bitmap);
d785a06a
N
360}
361
362static void end_bitmap_write(struct buffer_head *bh, int uptodate)
363{
364 struct bitmap *bitmap = bh->b_private;
365 unsigned long flags;
32a7627c 366
d785a06a
N
367 if (!uptodate) {
368 spin_lock_irqsave(&bitmap->lock, flags);
369 bitmap->flags |= BITMAP_WRITE_ERROR;
370 spin_unlock_irqrestore(&bitmap->lock, flags);
32a7627c 371 }
d785a06a
N
372 if (atomic_dec_and_test(&bitmap->pending_writes))
373 wake_up(&bitmap->write_wait);
374}
32a7627c 375
d785a06a
N
376/* copied from buffer.c */
377static void
378__clear_page_buffers(struct page *page)
379{
380 ClearPagePrivate(page);
381 set_page_private(page, 0);
382 page_cache_release(page);
383}
384static void free_buffers(struct page *page)
385{
386 struct buffer_head *bh = page_buffers(page);
77ad4bc7 387
d785a06a
N
388 while (bh) {
389 struct buffer_head *next = bh->b_this_page;
390 free_buffer_head(bh);
391 bh = next;
77ad4bc7 392 }
d785a06a
N
393 __clear_page_buffers(page);
394 put_page(page);
32a7627c
N
395}
396
d785a06a
N
397/* read a page from a file.
398 * We both read the page, and attach buffers to the page to record the
399 * address of each block (using bmap). These addresses will be used
400 * to write the block later, completely bypassing the filesystem.
401 * This usage is similar to how swap files are handled, and allows us
402 * to write to a file with no concerns of memory allocation failing.
403 */
32a7627c 404static struct page *read_page(struct file *file, unsigned long index,
d785a06a
N
405 struct bitmap *bitmap,
406 unsigned long count)
32a7627c 407{
32a7627c 408 struct page *page = NULL;
c649bb9c 409 struct inode *inode = file->f_path.dentry->d_inode;
d785a06a
N
410 struct buffer_head *bh;
411 sector_t block;
32a7627c 412
ac2f40be 413 PRINTK("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
2d1f3b5d 414 (unsigned long long)index << PAGE_SHIFT);
32a7627c 415
d785a06a
N
416 page = alloc_page(GFP_KERNEL);
417 if (!page)
418 page = ERR_PTR(-ENOMEM);
32a7627c
N
419 if (IS_ERR(page))
420 goto out;
d785a06a 421
d785a06a
N
422 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
423 if (!bh) {
2d1f3b5d 424 put_page(page);
d785a06a 425 page = ERR_PTR(-ENOMEM);
32a7627c
N
426 goto out;
427 }
d785a06a
N
428 attach_page_buffers(page, bh);
429 block = index << (PAGE_SHIFT - inode->i_blkbits);
430 while (bh) {
431 if (count == 0)
432 bh->b_blocknr = 0;
433 else {
434 bh->b_blocknr = bmap(inode, block);
435 if (bh->b_blocknr == 0) {
436 /* Cannot use this file! */
437 free_buffers(page);
438 page = ERR_PTR(-EINVAL);
439 goto out;
440 }
441 bh->b_bdev = inode->i_sb->s_bdev;
442 if (count < (1<<inode->i_blkbits))
443 count = 0;
444 else
445 count -= (1<<inode->i_blkbits);
446
447 bh->b_end_io = end_bitmap_write;
448 bh->b_private = bitmap;
ce25c31b
N
449 atomic_inc(&bitmap->pending_writes);
450 set_buffer_locked(bh);
451 set_buffer_mapped(bh);
452 submit_bh(READ, bh);
d785a06a
N
453 }
454 block++;
455 bh = bh->b_this_page;
456 }
d785a06a 457 page->index = index;
ce25c31b
N
458
459 wait_event(bitmap->write_wait,
460 atomic_read(&bitmap->pending_writes)==0);
461 if (bitmap->flags & BITMAP_WRITE_ERROR) {
462 free_buffers(page);
463 page = ERR_PTR(-EIO);
464 }
32a7627c
N
465out:
466 if (IS_ERR(page))
ac2f40be 467 printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %ld\n",
2d1f3b5d
N
468 (int)PAGE_SIZE,
469 (unsigned long long)index << PAGE_SHIFT,
32a7627c
N
470 PTR_ERR(page));
471 return page;
472}
473
474/*
475 * bitmap file superblock operations
476 */
477
478/* update the event counter and sync the superblock to disk */
4ad13663 479void bitmap_update_sb(struct bitmap *bitmap)
32a7627c
N
480{
481 bitmap_super_t *sb;
482 unsigned long flags;
483
484 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
4ad13663 485 return;
ece5cff0
N
486 if (bitmap->mddev->bitmap_info.external)
487 return;
32a7627c
N
488 spin_lock_irqsave(&bitmap->lock, flags);
489 if (!bitmap->sb_page) { /* no superblock */
490 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 491 return;
32a7627c 492 }
32a7627c 493 spin_unlock_irqrestore(&bitmap->lock, flags);
7b92813c 494 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 495 sb->events = cpu_to_le64(bitmap->mddev->events);
a0da84f3
NB
496 if (bitmap->mddev->events < bitmap->events_cleared) {
497 /* rocking back to read-only */
498 bitmap->events_cleared = bitmap->mddev->events;
499 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
500 }
43a70507
N
501 /* Just in case these have been changed via sysfs: */
502 sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
503 sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
ea03aff9 504 kunmap_atomic(sb, KM_USER0);
4ad13663 505 write_page(bitmap, bitmap->sb_page, 1);
32a7627c
N
506}
507
508/* print out the bitmap file superblock */
509void bitmap_print_sb(struct bitmap *bitmap)
510{
511 bitmap_super_t *sb;
512
513 if (!bitmap || !bitmap->sb_page)
514 return;
7b92813c 515 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 516 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
a2cff26a
N
517 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
518 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
519 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
32a7627c
N
520 *(__u32 *)(sb->uuid+0),
521 *(__u32 *)(sb->uuid+4),
522 *(__u32 *)(sb->uuid+8),
523 *(__u32 *)(sb->uuid+12));
a2cff26a 524 printk(KERN_DEBUG " events: %llu\n",
32a7627c 525 (unsigned long long) le64_to_cpu(sb->events));
a2cff26a 526 printk(KERN_DEBUG "events cleared: %llu\n",
32a7627c 527 (unsigned long long) le64_to_cpu(sb->events_cleared));
a2cff26a
N
528 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
529 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
530 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
531 printk(KERN_DEBUG " sync size: %llu KB\n",
532 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
4b6d287f 533 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
ea03aff9 534 kunmap_atomic(sb, KM_USER0);
32a7627c
N
535}
536
537/* read the superblock from the bitmap file and initialize some bitmap fields */
538static int bitmap_read_sb(struct bitmap *bitmap)
539{
540 char *reason = NULL;
541 bitmap_super_t *sb;
4b6d287f 542 unsigned long chunksize, daemon_sleep, write_behind;
32a7627c
N
543 unsigned long long events;
544 int err = -EINVAL;
545
546 /* page 0 is the superblock, read it... */
f49d5e62
N
547 if (bitmap->file) {
548 loff_t isize = i_size_read(bitmap->file->f_mapping->host);
549 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
550
551 bitmap->sb_page = read_page(bitmap->file, 0, bitmap, bytes);
552 } else {
42a04b50
N
553 bitmap->sb_page = read_sb_page(bitmap->mddev,
554 bitmap->mddev->bitmap_info.offset,
a2ed9615
N
555 NULL,
556 0, sizeof(bitmap_super_t));
a654b9d8 557 }
32a7627c
N
558 if (IS_ERR(bitmap->sb_page)) {
559 err = PTR_ERR(bitmap->sb_page);
560 bitmap->sb_page = NULL;
561 return err;
562 }
563
7b92813c 564 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 565
32a7627c 566 chunksize = le32_to_cpu(sb->chunksize);
1b04be96 567 daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
4b6d287f 568 write_behind = le32_to_cpu(sb->write_behind);
32a7627c
N
569
570 /* verify that the bitmap-specific fields are valid */
571 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
572 reason = "bad magic";
bd926c63
N
573 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
574 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
32a7627c 575 reason = "unrecognized superblock version";
1187cf0a 576 else if (chunksize < 512)
7dd5d34c 577 reason = "bitmap chunksize too small";
32a7627c
N
578 else if ((1 << ffz(~chunksize)) != chunksize)
579 reason = "bitmap chunksize not a power of 2";
1b04be96 580 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
7dd5d34c 581 reason = "daemon sleep period out of range";
4b6d287f
N
582 else if (write_behind > COUNTER_MAX)
583 reason = "write-behind limit out of range (0 - 16383)";
32a7627c
N
584 if (reason) {
585 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
586 bmname(bitmap), reason);
587 goto out;
588 }
589
590 /* keep the array size field of the bitmap superblock up to date */
591 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
592
593 if (!bitmap->mddev->persistent)
594 goto success;
595
596 /*
597 * if we have a persistent array superblock, compare the
598 * bitmap's UUID and event counter to the mddev's
599 */
600 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
601 printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n",
602 bmname(bitmap));
603 goto out;
604 }
605 events = le64_to_cpu(sb->events);
606 if (events < bitmap->mddev->events) {
607 printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) "
608 "-- forcing full recovery\n", bmname(bitmap), events,
609 (unsigned long long) bitmap->mddev->events);
4f2e639a 610 sb->state |= cpu_to_le32(BITMAP_STALE);
32a7627c
N
611 }
612success:
613 /* assign fields using values from superblock */
42a04b50
N
614 bitmap->mddev->bitmap_info.chunksize = chunksize;
615 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
42a04b50 616 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
4f2e639a 617 bitmap->flags |= le32_to_cpu(sb->state);
bd926c63
N
618 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
619 bitmap->flags |= BITMAP_HOSTENDIAN;
32a7627c 620 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
4f2e639a 621 if (sb->state & cpu_to_le32(BITMAP_STALE))
6a07997f 622 bitmap->events_cleared = bitmap->mddev->events;
32a7627c
N
623 err = 0;
624out:
ea03aff9 625 kunmap_atomic(sb, KM_USER0);
32a7627c
N
626 if (err)
627 bitmap_print_sb(bitmap);
628 return err;
629}
630
631enum bitmap_mask_op {
632 MASK_SET,
633 MASK_UNSET
634};
635
4ad13663
N
636/* record the state of the bitmap in the superblock. Return the old value */
637static int bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
638 enum bitmap_mask_op op)
32a7627c
N
639{
640 bitmap_super_t *sb;
641 unsigned long flags;
4ad13663 642 int old;
32a7627c
N
643
644 spin_lock_irqsave(&bitmap->lock, flags);
7e317655 645 if (!bitmap->sb_page) { /* can't set the state */
32a7627c 646 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 647 return 0;
32a7627c 648 }
32a7627c 649 spin_unlock_irqrestore(&bitmap->lock, flags);
7b92813c 650 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
4ad13663 651 old = le32_to_cpu(sb->state) & bits;
32a7627c 652 switch (op) {
ac2f40be
N
653 case MASK_SET:
654 sb->state |= cpu_to_le32(bits);
655 break;
656 case MASK_UNSET:
657 sb->state &= cpu_to_le32(~bits);
658 break;
659 default:
660 BUG();
32a7627c 661 }
ea03aff9 662 kunmap_atomic(sb, KM_USER0);
4ad13663 663 return old;
32a7627c
N
664}
665
666/*
667 * general bitmap file operations
668 */
669
ece5cff0
N
670/*
671 * on-disk bitmap:
672 *
673 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
674 * file a page at a time. There's a superblock at the start of the file.
675 */
32a7627c 676/* calculate the index of the page that contains this bit */
ece5cff0 677static inline unsigned long file_page_index(struct bitmap *bitmap, unsigned long chunk)
32a7627c 678{
ece5cff0
N
679 if (!bitmap->mddev->bitmap_info.external)
680 chunk += sizeof(bitmap_super_t) << 3;
681 return chunk >> PAGE_BIT_SHIFT;
32a7627c
N
682}
683
684/* calculate the (bit) offset of this bit within a page */
ece5cff0 685static inline unsigned long file_page_offset(struct bitmap *bitmap, unsigned long chunk)
32a7627c 686{
ece5cff0
N
687 if (!bitmap->mddev->bitmap_info.external)
688 chunk += sizeof(bitmap_super_t) << 3;
689 return chunk & (PAGE_BITS - 1);
32a7627c
N
690}
691
692/*
693 * return a pointer to the page in the filemap that contains the given bit
694 *
695 * this lookup is complicated by the fact that the bitmap sb might be exactly
696 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
697 * 0 or page 1
698 */
699static inline struct page *filemap_get_page(struct bitmap *bitmap,
700 unsigned long chunk)
701{
e384e585
N
702 if (bitmap->filemap == NULL)
703 return NULL;
ac2f40be
N
704 if (file_page_index(bitmap, chunk) >= bitmap->file_pages)
705 return NULL;
ece5cff0
N
706 return bitmap->filemap[file_page_index(bitmap, chunk)
707 - file_page_index(bitmap, 0)];
32a7627c
N
708}
709
32a7627c
N
710static void bitmap_file_unmap(struct bitmap *bitmap)
711{
712 struct page **map, *sb_page;
713 unsigned long *attr;
714 int pages;
715 unsigned long flags;
716
717 spin_lock_irqsave(&bitmap->lock, flags);
718 map = bitmap->filemap;
719 bitmap->filemap = NULL;
720 attr = bitmap->filemap_attr;
721 bitmap->filemap_attr = NULL;
722 pages = bitmap->file_pages;
723 bitmap->file_pages = 0;
724 sb_page = bitmap->sb_page;
725 bitmap->sb_page = NULL;
726 spin_unlock_irqrestore(&bitmap->lock, flags);
727
728 while (pages--)
ece5cff0 729 if (map[pages] != sb_page) /* 0 is sb_page, release it below */
d785a06a 730 free_buffers(map[pages]);
32a7627c
N
731 kfree(map);
732 kfree(attr);
733
d785a06a
N
734 if (sb_page)
735 free_buffers(sb_page);
32a7627c
N
736}
737
738static void bitmap_file_put(struct bitmap *bitmap)
739{
740 struct file *file;
32a7627c
N
741 unsigned long flags;
742
743 spin_lock_irqsave(&bitmap->lock, flags);
744 file = bitmap->file;
745 bitmap->file = NULL;
746 spin_unlock_irqrestore(&bitmap->lock, flags);
747
d785a06a
N
748 if (file)
749 wait_event(bitmap->write_wait,
750 atomic_read(&bitmap->pending_writes)==0);
32a7627c
N
751 bitmap_file_unmap(bitmap);
752
d785a06a 753 if (file) {
c649bb9c 754 struct inode *inode = file->f_path.dentry->d_inode;
fc0ecff6 755 invalidate_mapping_pages(inode->i_mapping, 0, -1);
32a7627c 756 fput(file);
d785a06a 757 }
32a7627c
N
758}
759
32a7627c
N
760/*
761 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
762 * then it is no longer reliable, so we stop using it and we mark the file
763 * as failed in the superblock
764 */
765static void bitmap_file_kick(struct bitmap *bitmap)
766{
767 char *path, *ptr = NULL;
768
4ad13663
N
769 if (bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET) == 0) {
770 bitmap_update_sb(bitmap);
32a7627c 771
4ad13663
N
772 if (bitmap->file) {
773 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
774 if (path)
6bcfd601
CH
775 ptr = d_path(&bitmap->file->f_path, path,
776 PAGE_SIZE);
777
4ad13663
N
778 printk(KERN_ALERT
779 "%s: kicking failed bitmap file %s from array!\n",
6bcfd601 780 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
32a7627c 781
4ad13663
N
782 kfree(path);
783 } else
784 printk(KERN_ALERT
785 "%s: disabling internal bitmap due to errors\n",
786 bmname(bitmap));
a654b9d8 787 }
32a7627c
N
788
789 bitmap_file_put(bitmap);
790
791 return;
792}
793
794enum bitmap_page_attr {
ac2f40be
N
795 BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */
796 BITMAP_PAGE_CLEAN = 1, /* there are bits that might need to be cleared */
797 BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
32a7627c
N
798};
799
800static inline void set_page_attr(struct bitmap *bitmap, struct page *page,
801 enum bitmap_page_attr attr)
802{
e384e585
N
803 if (page)
804 __set_bit((page->index<<2) + attr, bitmap->filemap_attr);
805 else
806 __set_bit(attr, &bitmap->logattrs);
32a7627c
N
807}
808
809static inline void clear_page_attr(struct bitmap *bitmap, struct page *page,
810 enum bitmap_page_attr attr)
811{
e384e585
N
812 if (page)
813 __clear_bit((page->index<<2) + attr, bitmap->filemap_attr);
814 else
815 __clear_bit(attr, &bitmap->logattrs);
32a7627c
N
816}
817
ec7a3197
N
818static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page,
819 enum bitmap_page_attr attr)
32a7627c 820{
e384e585
N
821 if (page)
822 return test_bit((page->index<<2) + attr, bitmap->filemap_attr);
823 else
824 return test_bit(attr, &bitmap->logattrs);
32a7627c
N
825}
826
827/*
828 * bitmap_file_set_bit -- called before performing a write to the md device
829 * to set (and eventually sync) a particular bit in the bitmap file
830 *
831 * we set the bit immediately, then we record the page number so that
832 * when an unplug occurs, we can flush the dirty pages out to disk
833 */
834static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
835{
836 unsigned long bit;
e384e585 837 struct page *page = NULL;
32a7627c
N
838 void *kaddr;
839 unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap);
840
e384e585
N
841 if (!bitmap->filemap) {
842 struct dm_dirty_log *log = bitmap->mddev->bitmap_info.log;
843 if (log)
844 log->type->mark_region(log, chunk);
845 } else {
32a7627c 846
e384e585
N
847 page = filemap_get_page(bitmap, chunk);
848 if (!page)
849 return;
850 bit = file_page_offset(bitmap, chunk);
32a7627c 851
e384e585
N
852 /* set the bit */
853 kaddr = kmap_atomic(page, KM_USER0);
854 if (bitmap->flags & BITMAP_HOSTENDIAN)
855 set_bit(bit, kaddr);
856 else
6b33aff3 857 __test_and_set_bit_le(bit, kaddr);
e384e585
N
858 kunmap_atomic(kaddr, KM_USER0);
859 PRINTK("set file bit %lu page %lu\n", bit, page->index);
860 }
32a7627c
N
861 /* record page number so it gets flushed to disk when unplug occurs */
862 set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
32a7627c
N
863}
864
865/* this gets called when the md device is ready to unplug its underlying
866 * (slave) device queues -- before we let any writes go down, we need to
867 * sync the dirty pages of the bitmap file to disk */
4ad13663 868void bitmap_unplug(struct bitmap *bitmap)
32a7627c 869{
ec7a3197
N
870 unsigned long i, flags;
871 int dirty, need_write;
32a7627c
N
872 struct page *page;
873 int wait = 0;
874
875 if (!bitmap)
4ad13663 876 return;
e384e585
N
877 if (!bitmap->filemap) {
878 /* Must be using a dirty_log */
879 struct dm_dirty_log *log = bitmap->mddev->bitmap_info.log;
880 dirty = test_and_clear_bit(BITMAP_PAGE_DIRTY, &bitmap->logattrs);
881 need_write = test_and_clear_bit(BITMAP_PAGE_NEEDWRITE, &bitmap->logattrs);
882 if (dirty || need_write)
883 if (log->type->flush(log))
884 bitmap->flags |= BITMAP_WRITE_ERROR;
885 goto out;
886 }
32a7627c
N
887
888 /* look at each page to see if there are any set bits that need to be
889 * flushed out to disk */
890 for (i = 0; i < bitmap->file_pages; i++) {
891 spin_lock_irqsave(&bitmap->lock, flags);
a654b9d8 892 if (!bitmap->filemap) {
32a7627c 893 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 894 return;
32a7627c
N
895 }
896 page = bitmap->filemap[i];
ec7a3197
N
897 dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
898 need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
32a7627c
N
899 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
900 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
ec7a3197 901 if (dirty)
32a7627c
N
902 wait = 1;
903 spin_unlock_irqrestore(&bitmap->lock, flags);
904
ac2f40be 905 if (dirty || need_write)
4ad13663 906 write_page(bitmap, page, 0);
32a7627c
N
907 }
908 if (wait) { /* if any writes were performed, we need to wait on them */
0b79ccf0 909 if (bitmap->file)
d785a06a
N
910 wait_event(bitmap->write_wait,
911 atomic_read(&bitmap->pending_writes)==0);
0b79ccf0 912 else
a9701a30 913 md_super_wait(bitmap->mddev);
32a7627c 914 }
e384e585 915out:
d785a06a
N
916 if (bitmap->flags & BITMAP_WRITE_ERROR)
917 bitmap_file_kick(bitmap);
32a7627c 918}
ac2f40be 919EXPORT_SYMBOL(bitmap_unplug);
32a7627c 920
6a07997f 921static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
32a7627c
N
922/* * bitmap_init_from_disk -- called at bitmap_create time to initialize
923 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
924 * memory mapping of the bitmap file
925 * Special cases:
926 * if there's no bitmap file, or if the bitmap file had been
927 * previously kicked from the array, we mark all the bits as
928 * 1's in order to cause a full resync.
6a07997f
N
929 *
930 * We ignore all bits for sectors that end earlier than 'start'.
931 * This is used when reading an out-of-date bitmap...
32a7627c 932 */
6a07997f 933static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
32a7627c
N
934{
935 unsigned long i, chunks, index, oldindex, bit;
936 struct page *page = NULL, *oldpage = NULL;
937 unsigned long num_pages, bit_cnt = 0;
938 struct file *file;
d785a06a 939 unsigned long bytes, offset;
32a7627c
N
940 int outofdate;
941 int ret = -ENOSPC;
ea03aff9 942 void *paddr;
32a7627c
N
943
944 chunks = bitmap->chunks;
945 file = bitmap->file;
946
42a04b50 947 BUG_ON(!file && !bitmap->mddev->bitmap_info.offset);
32a7627c 948
44456d37 949#ifdef INJECT_FAULTS_3
32a7627c
N
950 outofdate = 1;
951#else
952 outofdate = bitmap->flags & BITMAP_STALE;
953#endif
954 if (outofdate)
955 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
956 "recovery\n", bmname(bitmap));
957
e384e585 958 bytes = DIV_ROUND_UP(bitmap->chunks, 8);
ece5cff0
N
959 if (!bitmap->mddev->bitmap_info.external)
960 bytes += sizeof(bitmap_super_t);
bc7f77de 961
e384e585 962 num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
bc7f77de 963
ece5cff0 964 if (file && i_size_read(file->f_mapping->host) < bytes) {
32a7627c
N
965 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
966 bmname(bitmap),
967 (unsigned long) i_size_read(file->f_mapping->host),
ece5cff0 968 bytes);
4ad13663 969 goto err;
32a7627c 970 }
bc7f77de
N
971
972 ret = -ENOMEM;
973
32a7627c 974 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
bc7f77de 975 if (!bitmap->filemap)
4ad13663 976 goto err;
32a7627c 977
e16b68b6
N
978 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
979 bitmap->filemap_attr = kzalloc(
ac2f40be 980 roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
e16b68b6 981 GFP_KERNEL);
bc7f77de 982 if (!bitmap->filemap_attr)
4ad13663 983 goto err;
32a7627c 984
32a7627c
N
985 oldindex = ~0L;
986
987 for (i = 0; i < chunks; i++) {
bd926c63 988 int b;
ece5cff0
N
989 index = file_page_index(bitmap, i);
990 bit = file_page_offset(bitmap, i);
32a7627c 991 if (index != oldindex) { /* this is a new page, read it in */
d785a06a 992 int count;
32a7627c 993 /* unmap the old page, we're done with it */
d785a06a 994 if (index == num_pages-1)
ece5cff0 995 count = bytes - index * PAGE_SIZE;
d785a06a
N
996 else
997 count = PAGE_SIZE;
ece5cff0 998 if (index == 0 && bitmap->sb_page) {
32a7627c
N
999 /*
1000 * if we're here then the superblock page
1001 * contains some bits (PAGE_SIZE != sizeof sb)
1002 * we've already read it in, so just use it
1003 */
1004 page = bitmap->sb_page;
1005 offset = sizeof(bitmap_super_t);
53845270 1006 if (!file)
5c04f551
VK
1007 page = read_sb_page(
1008 bitmap->mddev,
1009 bitmap->mddev->bitmap_info.offset,
1010 page,
1011 index, count);
a654b9d8 1012 } else if (file) {
d785a06a 1013 page = read_page(file, index, bitmap, count);
a654b9d8
N
1014 offset = 0;
1015 } else {
42a04b50
N
1016 page = read_sb_page(bitmap->mddev,
1017 bitmap->mddev->bitmap_info.offset,
a2ed9615
N
1018 NULL,
1019 index, count);
32a7627c
N
1020 offset = 0;
1021 }
a654b9d8
N
1022 if (IS_ERR(page)) { /* read error */
1023 ret = PTR_ERR(page);
4ad13663 1024 goto err;
a654b9d8
N
1025 }
1026
32a7627c
N
1027 oldindex = index;
1028 oldpage = page;
32a7627c 1029
b74fd282
N
1030 bitmap->filemap[bitmap->file_pages++] = page;
1031 bitmap->last_page_size = count;
1032
32a7627c
N
1033 if (outofdate) {
1034 /*
1035 * if bitmap is out of date, dirty the
ac2f40be 1036 * whole page and write it out
32a7627c 1037 */
ea03aff9
N
1038 paddr = kmap_atomic(page, KM_USER0);
1039 memset(paddr + offset, 0xff,
6a07997f 1040 PAGE_SIZE - offset);
ea03aff9 1041 kunmap_atomic(paddr, KM_USER0);
4ad13663
N
1042 write_page(bitmap, page, 1);
1043
1044 ret = -EIO;
b74fd282 1045 if (bitmap->flags & BITMAP_WRITE_ERROR)
4ad13663 1046 goto err;
32a7627c 1047 }
32a7627c 1048 }
ea03aff9 1049 paddr = kmap_atomic(page, KM_USER0);
bd926c63 1050 if (bitmap->flags & BITMAP_HOSTENDIAN)
ea03aff9 1051 b = test_bit(bit, paddr);
bd926c63 1052 else
6b33aff3 1053 b = test_bit_le(bit, paddr);
ea03aff9 1054 kunmap_atomic(paddr, KM_USER0);
bd926c63 1055 if (b) {
32a7627c 1056 /* if the disk bit is set, set the memory bit */
db305e50
N
1057 int needed = ((sector_t)(i+1) << (CHUNK_BLOCK_SHIFT(bitmap))
1058 >= start);
1059 bitmap_set_memory_bits(bitmap,
1060 (sector_t)i << CHUNK_BLOCK_SHIFT(bitmap),
1061 needed);
32a7627c 1062 bit_cnt++;
6a07997f 1063 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
32a7627c 1064 }
32a7627c
N
1065 }
1066
ac2f40be 1067 /* everything went OK */
32a7627c
N
1068 ret = 0;
1069 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
1070
32a7627c
N
1071 if (bit_cnt) { /* Kick recovery if any bits were set */
1072 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1073 md_wakeup_thread(bitmap->mddev->thread);
1074 }
1075
32a7627c 1076 printk(KERN_INFO "%s: bitmap initialized from disk: "
4ad13663
N
1077 "read %lu/%lu pages, set %lu bits\n",
1078 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt);
1079
1080 return 0;
32a7627c 1081
4ad13663
N
1082 err:
1083 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1084 bmname(bitmap), ret);
32a7627c
N
1085 return ret;
1086}
1087
a654b9d8
N
1088void bitmap_write_all(struct bitmap *bitmap)
1089{
1090 /* We don't actually write all bitmap blocks here,
1091 * just flag them as needing to be written
1092 */
ec7a3197 1093 int i;
a654b9d8 1094
ac2f40be 1095 for (i = 0; i < bitmap->file_pages; i++)
ec7a3197
N
1096 set_page_attr(bitmap, bitmap->filemap[i],
1097 BITMAP_PAGE_NEEDWRITE);
a654b9d8
N
1098}
1099
32a7627c
N
1100static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
1101{
1102 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1103 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1104 bitmap->bp[page].count += inc;
32a7627c
N
1105 bitmap_checkfree(bitmap, page);
1106}
1107static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
57dab0bd 1108 sector_t offset, sector_t *blocks,
32a7627c
N
1109 int create);
1110
1111/*
1112 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1113 * out to disk
1114 */
1115
aa5cbd10 1116void bitmap_daemon_work(mddev_t *mddev)
32a7627c 1117{
aa5cbd10 1118 struct bitmap *bitmap;
aa3163f8 1119 unsigned long j;
32a7627c
N
1120 unsigned long flags;
1121 struct page *page = NULL, *lastpage = NULL;
57dab0bd 1122 sector_t blocks;
ea03aff9 1123 void *paddr;
e384e585 1124 struct dm_dirty_log *log = mddev->bitmap_info.log;
32a7627c 1125
aa5cbd10
N
1126 /* Use a mutex to guard daemon_work against
1127 * bitmap_destroy.
1128 */
c3d9714e 1129 mutex_lock(&mddev->bitmap_info.mutex);
aa5cbd10
N
1130 bitmap = mddev->bitmap;
1131 if (bitmap == NULL) {
c3d9714e 1132 mutex_unlock(&mddev->bitmap_info.mutex);
4ad13663 1133 return;
aa5cbd10 1134 }
42a04b50 1135 if (time_before(jiffies, bitmap->daemon_lastrun
1b04be96 1136 + bitmap->mddev->bitmap_info.daemon_sleep))
7be3dfec
N
1137 goto done;
1138
32a7627c 1139 bitmap->daemon_lastrun = jiffies;
8311c29d
N
1140 if (bitmap->allclean) {
1141 bitmap->mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
aa5cbd10 1142 goto done;
8311c29d
N
1143 }
1144 bitmap->allclean = 1;
32a7627c 1145
be512691 1146 spin_lock_irqsave(&bitmap->lock, flags);
32a7627c
N
1147 for (j = 0; j < bitmap->chunks; j++) {
1148 bitmap_counter_t *bmc;
e384e585
N
1149 if (!bitmap->filemap) {
1150 if (!log)
1151 /* error or shutdown */
1152 break;
1153 } else
1154 page = filemap_get_page(bitmap, j);
32a7627c
N
1155
1156 if (page != lastpage) {
aa3163f8 1157 /* skip this page unless it's marked as needing cleaning */
ec7a3197
N
1158 if (!test_page_attr(bitmap, page, BITMAP_PAGE_CLEAN)) {
1159 int need_write = test_page_attr(bitmap, page,
1160 BITMAP_PAGE_NEEDWRITE);
a647e4bc 1161 if (need_write)
aa3163f8 1162 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
a647e4bc 1163
aa3163f8 1164 spin_unlock_irqrestore(&bitmap->lock, flags);
8311c29d 1165 if (need_write) {
4ad13663 1166 write_page(bitmap, page, 0);
8311c29d
N
1167 bitmap->allclean = 0;
1168 }
be512691
N
1169 spin_lock_irqsave(&bitmap->lock, flags);
1170 j |= (PAGE_BITS - 1);
aa3163f8
N
1171 continue;
1172 }
1173
32a7627c 1174 /* grab the new page, sync and release the old */
32a7627c 1175 if (lastpage != NULL) {
ec7a3197 1176 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
32a7627c
N
1177 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1178 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 1179 write_page(bitmap, lastpage, 0);
32a7627c
N
1180 } else {
1181 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1182 spin_unlock_irqrestore(&bitmap->lock, flags);
1183 }
32a7627c
N
1184 } else
1185 spin_unlock_irqrestore(&bitmap->lock, flags);
1186 lastpage = page;
a0da84f3
NB
1187
1188 /* We are possibly going to clear some bits, so make
1189 * sure that events_cleared is up-to-date.
1190 */
ece5cff0
N
1191 if (bitmap->need_sync &&
1192 bitmap->mddev->bitmap_info.external == 0) {
a0da84f3
NB
1193 bitmap_super_t *sb;
1194 bitmap->need_sync = 0;
1195 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
1196 sb->events_cleared =
1197 cpu_to_le64(bitmap->events_cleared);
1198 kunmap_atomic(sb, KM_USER0);
1199 write_page(bitmap, bitmap->sb_page, 1);
1200 }
32a7627c 1201 spin_lock_irqsave(&bitmap->lock, flags);
ece5cff0
N
1202 if (!bitmap->need_sync)
1203 clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
32a7627c 1204 }
db305e50
N
1205 bmc = bitmap_get_counter(bitmap,
1206 (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap),
1207 &blocks, 0);
32a7627c 1208 if (bmc) {
8311c29d
N
1209 if (*bmc)
1210 bitmap->allclean = 0;
1211
32a7627c 1212 if (*bmc == 2) {
ac2f40be 1213 *bmc = 1; /* maybe clear the bit next time */
32a7627c 1214 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
ece5cff0 1215 } else if (*bmc == 1 && !bitmap->need_sync) {
32a7627c
N
1216 /* we can clear the bit */
1217 *bmc = 0;
db305e50
N
1218 bitmap_count_page(bitmap,
1219 (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap),
32a7627c
N
1220 -1);
1221
1222 /* clear the bit */
e384e585
N
1223 if (page) {
1224 paddr = kmap_atomic(page, KM_USER0);
1225 if (bitmap->flags & BITMAP_HOSTENDIAN)
1226 clear_bit(file_page_offset(bitmap, j),
1227 paddr);
1228 else
6b33aff3 1229 __test_and_clear_bit_le(file_page_offset(bitmap, j),
e384e585
N
1230 paddr);
1231 kunmap_atomic(paddr, KM_USER0);
1232 } else
1233 log->type->clear_region(log, j);
32a7627c 1234 }
be512691
N
1235 } else
1236 j |= PAGE_COUNTER_MASK;
32a7627c 1237 }
be512691 1238 spin_unlock_irqrestore(&bitmap->lock, flags);
32a7627c
N
1239
1240 /* now sync the final page */
e384e585 1241 if (lastpage != NULL || log != NULL) {
32a7627c 1242 spin_lock_irqsave(&bitmap->lock, flags);
ec7a3197 1243 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
32a7627c
N
1244 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1245 spin_unlock_irqrestore(&bitmap->lock, flags);
e384e585
N
1246 if (lastpage)
1247 write_page(bitmap, lastpage, 0);
1248 else
1249 if (log->type->flush(log))
1250 bitmap->flags |= BITMAP_WRITE_ERROR;
32a7627c
N
1251 } else {
1252 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1253 spin_unlock_irqrestore(&bitmap->lock, flags);
1254 }
32a7627c
N
1255 }
1256
7be3dfec 1257 done:
8311c29d 1258 if (bitmap->allclean == 0)
ac2f40be 1259 bitmap->mddev->thread->timeout =
1b04be96 1260 bitmap->mddev->bitmap_info.daemon_sleep;
c3d9714e 1261 mutex_unlock(&mddev->bitmap_info.mutex);
32a7627c
N
1262}
1263
32a7627c 1264static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
57dab0bd 1265 sector_t offset, sector_t *blocks,
32a7627c 1266 int create)
ee305ace
N
1267__releases(bitmap->lock)
1268__acquires(bitmap->lock)
32a7627c
N
1269{
1270 /* If 'create', we might release the lock and reclaim it.
1271 * The lock must have been taken with interrupts enabled.
1272 * If !create, we don't release the lock.
1273 */
1274 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1275 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1276 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1277 sector_t csize;
ef425673 1278 int err;
32a7627c 1279
ef425673
N
1280 err = bitmap_checkpage(bitmap, page, create);
1281
1282 if (bitmap->bp[page].hijacked ||
1283 bitmap->bp[page].map == NULL)
1284 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) +
1285 PAGE_COUNTER_SHIFT - 1);
1286 else
32a7627c 1287 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
ef425673
N
1288 *blocks = csize - (offset & (csize - 1));
1289
1290 if (err < 0)
32a7627c 1291 return NULL;
ef425673 1292
32a7627c
N
1293 /* now locked ... */
1294
1295 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1296 /* should we use the first or second counter field
1297 * of the hijacked pointer? */
1298 int hi = (pageoff > PAGE_COUNTER_MASK);
32a7627c
N
1299 return &((bitmap_counter_t *)
1300 &bitmap->bp[page].map)[hi];
ef425673 1301 } else /* page is allocated */
32a7627c
N
1302 return (bitmap_counter_t *)
1303 &(bitmap->bp[page].map[pageoff]);
32a7627c
N
1304}
1305
4b6d287f 1306int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
32a7627c 1307{
ac2f40be
N
1308 if (!bitmap)
1309 return 0;
4b6d287f
N
1310
1311 if (behind) {
696fcd53 1312 int bw;
4b6d287f 1313 atomic_inc(&bitmap->behind_writes);
696fcd53
PC
1314 bw = atomic_read(&bitmap->behind_writes);
1315 if (bw > bitmap->behind_writes_used)
1316 bitmap->behind_writes_used = bw;
1317
4b6d287f 1318 PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n",
696fcd53 1319 bw, bitmap->max_write_behind);
4b6d287f
N
1320 }
1321
32a7627c 1322 while (sectors) {
57dab0bd 1323 sector_t blocks;
32a7627c
N
1324 bitmap_counter_t *bmc;
1325
1326 spin_lock_irq(&bitmap->lock);
1327 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1328 if (!bmc) {
1329 spin_unlock_irq(&bitmap->lock);
1330 return 0;
1331 }
1332
da6e1a32
NB
1333 if (unlikely((*bmc & COUNTER_MAX) == COUNTER_MAX)) {
1334 DEFINE_WAIT(__wait);
1335 /* note that it is safe to do the prepare_to_wait
1336 * after the test as long as we do it before dropping
1337 * the spinlock.
1338 */
1339 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1340 TASK_UNINTERRUPTIBLE);
1341 spin_unlock_irq(&bitmap->lock);
7eaceacc 1342 io_schedule();
da6e1a32
NB
1343 finish_wait(&bitmap->overflow_wait, &__wait);
1344 continue;
1345 }
1346
ac2f40be 1347 switch (*bmc) {
32a7627c
N
1348 case 0:
1349 bitmap_file_set_bit(bitmap, offset);
ac2f40be 1350 bitmap_count_page(bitmap, offset, 1);
32a7627c
N
1351 /* fall through */
1352 case 1:
1353 *bmc = 2;
1354 }
da6e1a32 1355
32a7627c
N
1356 (*bmc)++;
1357
1358 spin_unlock_irq(&bitmap->lock);
1359
1360 offset += blocks;
1361 if (sectors > blocks)
1362 sectors -= blocks;
ac2f40be
N
1363 else
1364 sectors = 0;
32a7627c 1365 }
8311c29d 1366 bitmap->allclean = 0;
32a7627c
N
1367 return 0;
1368}
ac2f40be 1369EXPORT_SYMBOL(bitmap_startwrite);
32a7627c
N
1370
1371void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
4b6d287f 1372 int success, int behind)
32a7627c 1373{
ac2f40be
N
1374 if (!bitmap)
1375 return;
4b6d287f 1376 if (behind) {
e555190d
N
1377 if (atomic_dec_and_test(&bitmap->behind_writes))
1378 wake_up(&bitmap->behind_wait);
4b6d287f
N
1379 PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n",
1380 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1381 }
d0a4bb49
N
1382 if (bitmap->mddev->degraded)
1383 /* Never clear bits or update events_cleared when degraded */
1384 success = 0;
4b6d287f 1385
32a7627c 1386 while (sectors) {
57dab0bd 1387 sector_t blocks;
32a7627c
N
1388 unsigned long flags;
1389 bitmap_counter_t *bmc;
1390
1391 spin_lock_irqsave(&bitmap->lock, flags);
1392 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1393 if (!bmc) {
1394 spin_unlock_irqrestore(&bitmap->lock, flags);
1395 return;
1396 }
1397
a0da84f3
NB
1398 if (success &&
1399 bitmap->events_cleared < bitmap->mddev->events) {
1400 bitmap->events_cleared = bitmap->mddev->events;
1401 bitmap->need_sync = 1;
5ff5afff 1402 sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
a0da84f3
NB
1403 }
1404
32a7627c
N
1405 if (!success && ! (*bmc & NEEDED_MASK))
1406 *bmc |= NEEDED_MASK;
1407
da6e1a32
NB
1408 if ((*bmc & COUNTER_MAX) == COUNTER_MAX)
1409 wake_up(&bitmap->overflow_wait);
1410
32a7627c 1411 (*bmc)--;
ac2f40be 1412 if (*bmc <= 2)
32a7627c 1413 set_page_attr(bitmap,
e384e585
N
1414 filemap_get_page(
1415 bitmap,
1416 offset >> CHUNK_BLOCK_SHIFT(bitmap)),
32a7627c 1417 BITMAP_PAGE_CLEAN);
ac2f40be 1418
32a7627c
N
1419 spin_unlock_irqrestore(&bitmap->lock, flags);
1420 offset += blocks;
1421 if (sectors > blocks)
1422 sectors -= blocks;
ac2f40be
N
1423 else
1424 sectors = 0;
32a7627c
N
1425 }
1426}
ac2f40be 1427EXPORT_SYMBOL(bitmap_endwrite);
32a7627c 1428
57dab0bd 1429static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1187cf0a 1430 int degraded)
32a7627c
N
1431{
1432 bitmap_counter_t *bmc;
1433 int rv;
1434 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1435 *blocks = 1024;
1436 return 1; /* always resync if no bitmap */
1437 }
1438 spin_lock_irq(&bitmap->lock);
1439 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1440 rv = 0;
1441 if (bmc) {
1442 /* locked */
1443 if (RESYNC(*bmc))
1444 rv = 1;
1445 else if (NEEDED(*bmc)) {
1446 rv = 1;
6a806c51
N
1447 if (!degraded) { /* don't set/clear bits if degraded */
1448 *bmc |= RESYNC_MASK;
1449 *bmc &= ~NEEDED_MASK;
1450 }
32a7627c
N
1451 }
1452 }
1453 spin_unlock_irq(&bitmap->lock);
8311c29d 1454 bitmap->allclean = 0;
32a7627c
N
1455 return rv;
1456}
1457
57dab0bd 1458int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1187cf0a
N
1459 int degraded)
1460{
1461 /* bitmap_start_sync must always report on multiples of whole
1462 * pages, otherwise resync (which is very PAGE_SIZE based) will
1463 * get confused.
1464 * So call __bitmap_start_sync repeatedly (if needed) until
1465 * At least PAGE_SIZE>>9 blocks are covered.
1466 * Return the 'or' of the result.
1467 */
1468 int rv = 0;
57dab0bd 1469 sector_t blocks1;
1187cf0a
N
1470
1471 *blocks = 0;
1472 while (*blocks < (PAGE_SIZE>>9)) {
1473 rv |= __bitmap_start_sync(bitmap, offset,
1474 &blocks1, degraded);
1475 offset += blocks1;
1476 *blocks += blocks1;
1477 }
1478 return rv;
1479}
ac2f40be 1480EXPORT_SYMBOL(bitmap_start_sync);
1187cf0a 1481
57dab0bd 1482void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
32a7627c
N
1483{
1484 bitmap_counter_t *bmc;
1485 unsigned long flags;
ac2f40be
N
1486
1487 if (bitmap == NULL) {
32a7627c
N
1488 *blocks = 1024;
1489 return;
1490 }
1491 spin_lock_irqsave(&bitmap->lock, flags);
1492 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1493 if (bmc == NULL)
1494 goto unlock;
1495 /* locked */
32a7627c
N
1496 if (RESYNC(*bmc)) {
1497 *bmc &= ~RESYNC_MASK;
1498
1499 if (!NEEDED(*bmc) && aborted)
1500 *bmc |= NEEDED_MASK;
1501 else {
ac2f40be 1502 if (*bmc <= 2)
32a7627c
N
1503 set_page_attr(bitmap,
1504 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1505 BITMAP_PAGE_CLEAN);
32a7627c
N
1506 }
1507 }
1508 unlock:
1509 spin_unlock_irqrestore(&bitmap->lock, flags);
8311c29d 1510 bitmap->allclean = 0;
32a7627c 1511}
ac2f40be 1512EXPORT_SYMBOL(bitmap_end_sync);
32a7627c
N
1513
1514void bitmap_close_sync(struct bitmap *bitmap)
1515{
1516 /* Sync has finished, and any bitmap chunks that weren't synced
1517 * properly have been aborted. It remains to us to clear the
1518 * RESYNC bit wherever it is still on
1519 */
1520 sector_t sector = 0;
57dab0bd 1521 sector_t blocks;
b47490c9
N
1522 if (!bitmap)
1523 return;
32a7627c
N
1524 while (sector < bitmap->mddev->resync_max_sectors) {
1525 bitmap_end_sync(bitmap, sector, &blocks, 0);
b47490c9
N
1526 sector += blocks;
1527 }
1528}
ac2f40be 1529EXPORT_SYMBOL(bitmap_close_sync);
b47490c9
N
1530
1531void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector)
1532{
1533 sector_t s = 0;
57dab0bd 1534 sector_t blocks;
b47490c9
N
1535
1536 if (!bitmap)
1537 return;
1538 if (sector == 0) {
1539 bitmap->last_end_sync = jiffies;
1540 return;
1541 }
1542 if (time_before(jiffies, (bitmap->last_end_sync
1b04be96 1543 + bitmap->mddev->bitmap_info.daemon_sleep)))
b47490c9
N
1544 return;
1545 wait_event(bitmap->mddev->recovery_wait,
1546 atomic_read(&bitmap->mddev->recovery_active) == 0);
1547
75d3da43 1548 bitmap->mddev->curr_resync_completed = sector;
070dc6dd 1549 set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags);
b47490c9
N
1550 sector &= ~((1ULL << CHUNK_BLOCK_SHIFT(bitmap)) - 1);
1551 s = 0;
1552 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1553 bitmap_end_sync(bitmap, s, &blocks, 0);
1554 s += blocks;
32a7627c 1555 }
b47490c9 1556 bitmap->last_end_sync = jiffies;
acb180b0 1557 sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
32a7627c 1558}
ac2f40be 1559EXPORT_SYMBOL(bitmap_cond_end_sync);
32a7627c 1560
6a07997f 1561static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
32a7627c
N
1562{
1563 /* For each chunk covered by any of these sectors, set the
193f1c93 1564 * counter to 1 and set resync_needed. They should all
32a7627c
N
1565 * be 0 at this point
1566 */
193f1c93 1567
57dab0bd 1568 sector_t secs;
193f1c93
N
1569 bitmap_counter_t *bmc;
1570 spin_lock_irq(&bitmap->lock);
1571 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1572 if (!bmc) {
32a7627c 1573 spin_unlock_irq(&bitmap->lock);
193f1c93 1574 return;
32a7627c 1575 }
ac2f40be 1576 if (!*bmc) {
193f1c93 1577 struct page *page;
ac2f40be 1578 *bmc = 1 | (needed ? NEEDED_MASK : 0);
193f1c93
N
1579 bitmap_count_page(bitmap, offset, 1);
1580 page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap));
1581 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1582 }
1583 spin_unlock_irq(&bitmap->lock);
8311c29d 1584 bitmap->allclean = 0;
32a7627c
N
1585}
1586
9b1d1dac
PC
1587/* dirty the memory and file bits for bitmap chunks "s" to "e" */
1588void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1589{
1590 unsigned long chunk;
1591
1592 for (chunk = s; chunk <= e; chunk++) {
db305e50 1593 sector_t sec = (sector_t)chunk << CHUNK_BLOCK_SHIFT(bitmap);
9b1d1dac
PC
1594 bitmap_set_memory_bits(bitmap, sec, 1);
1595 bitmap_file_set_bit(bitmap, sec);
ffa23322
N
1596 if (sec < bitmap->mddev->recovery_cp)
1597 /* We are asserting that the array is dirty,
1598 * so move the recovery_cp address back so
1599 * that it is obvious that it is dirty
1600 */
1601 bitmap->mddev->recovery_cp = sec;
9b1d1dac
PC
1602 }
1603}
1604
6b8b3e8a
N
1605/*
1606 * flush out any pending updates
1607 */
1608void bitmap_flush(mddev_t *mddev)
1609{
1610 struct bitmap *bitmap = mddev->bitmap;
42a04b50 1611 long sleep;
6b8b3e8a
N
1612
1613 if (!bitmap) /* there was no bitmap */
1614 return;
1615
1616 /* run the daemon_work three time to ensure everything is flushed
1617 * that can be
1618 */
1b04be96 1619 sleep = mddev->bitmap_info.daemon_sleep * 2;
42a04b50 1620 bitmap->daemon_lastrun -= sleep;
aa5cbd10 1621 bitmap_daemon_work(mddev);
42a04b50 1622 bitmap->daemon_lastrun -= sleep;
aa5cbd10 1623 bitmap_daemon_work(mddev);
42a04b50 1624 bitmap->daemon_lastrun -= sleep;
aa5cbd10 1625 bitmap_daemon_work(mddev);
6b8b3e8a
N
1626 bitmap_update_sb(bitmap);
1627}
1628
32a7627c
N
1629/*
1630 * free memory that was allocated
1631 */
3178b0db 1632static void bitmap_free(struct bitmap *bitmap)
32a7627c
N
1633{
1634 unsigned long k, pages;
1635 struct bitmap_page *bp;
32a7627c
N
1636
1637 if (!bitmap) /* there was no bitmap */
1638 return;
1639
32a7627c
N
1640 /* release the bitmap file and kill the daemon */
1641 bitmap_file_put(bitmap);
1642
1643 bp = bitmap->bp;
1644 pages = bitmap->pages;
1645
1646 /* free all allocated memory */
1647
32a7627c
N
1648 if (bp) /* deallocate the page memory */
1649 for (k = 0; k < pages; k++)
1650 if (bp[k].map && !bp[k].hijacked)
1651 kfree(bp[k].map);
1652 kfree(bp);
1653 kfree(bitmap);
1654}
aa5cbd10 1655
3178b0db
N
1656void bitmap_destroy(mddev_t *mddev)
1657{
1658 struct bitmap *bitmap = mddev->bitmap;
1659
1660 if (!bitmap) /* there was no bitmap */
1661 return;
1662
c3d9714e 1663 mutex_lock(&mddev->bitmap_info.mutex);
3178b0db 1664 mddev->bitmap = NULL; /* disconnect from the md device */
c3d9714e 1665 mutex_unlock(&mddev->bitmap_info.mutex);
b15c2e57
N
1666 if (mddev->thread)
1667 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
3178b0db 1668
ece5cff0
N
1669 if (bitmap->sysfs_can_clear)
1670 sysfs_put(bitmap->sysfs_can_clear);
1671
3178b0db
N
1672 bitmap_free(bitmap);
1673}
32a7627c
N
1674
1675/*
1676 * initialize the bitmap structure
1677 * if this returns an error, bitmap_destroy must be called to do clean up
1678 */
1679int bitmap_create(mddev_t *mddev)
1680{
1681 struct bitmap *bitmap;
1f593903 1682 sector_t blocks = mddev->resync_max_sectors;
32a7627c
N
1683 unsigned long chunks;
1684 unsigned long pages;
c3d9714e 1685 struct file *file = mddev->bitmap_info.file;
32a7627c 1686 int err;
5ff5afff 1687 struct sysfs_dirent *bm = NULL;
32a7627c 1688
5f6e3c83 1689 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
32a7627c 1690
e384e585
N
1691 if (!file
1692 && !mddev->bitmap_info.offset
1693 && !mddev->bitmap_info.log) /* bitmap disabled, nothing to do */
32a7627c
N
1694 return 0;
1695
c3d9714e 1696 BUG_ON(file && mddev->bitmap_info.offset);
e384e585 1697 BUG_ON(mddev->bitmap_info.offset && mddev->bitmap_info.log);
a654b9d8 1698
9ffae0cf 1699 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
32a7627c
N
1700 if (!bitmap)
1701 return -ENOMEM;
1702
32a7627c 1703 spin_lock_init(&bitmap->lock);
ce25c31b
N
1704 atomic_set(&bitmap->pending_writes, 0);
1705 init_waitqueue_head(&bitmap->write_wait);
da6e1a32 1706 init_waitqueue_head(&bitmap->overflow_wait);
e555190d 1707 init_waitqueue_head(&bitmap->behind_wait);
ce25c31b 1708
32a7627c 1709 bitmap->mddev = mddev;
32a7627c 1710
5ff5afff
N
1711 if (mddev->kobj.sd)
1712 bm = sysfs_get_dirent(mddev->kobj.sd, NULL, "bitmap");
ece5cff0 1713 if (bm) {
3ff195b0 1714 bitmap->sysfs_can_clear = sysfs_get_dirent(bm, NULL, "can_clear");
ece5cff0
N
1715 sysfs_put(bm);
1716 } else
1717 bitmap->sysfs_can_clear = NULL;
1718
32a7627c 1719 bitmap->file = file;
ce25c31b
N
1720 if (file) {
1721 get_file(file);
ae8fa283
N
1722 /* As future accesses to this file will use bmap,
1723 * and bypass the page cache, we must sync the file
1724 * first.
1725 */
8018ab05 1726 vfs_fsync(file, 1);
ce25c31b 1727 }
42a04b50 1728 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
ece5cff0
N
1729 if (!mddev->bitmap_info.external)
1730 err = bitmap_read_sb(bitmap);
1731 else {
1732 err = 0;
1733 if (mddev->bitmap_info.chunksize == 0 ||
1734 mddev->bitmap_info.daemon_sleep == 0)
1735 /* chunksize and time_base need to be
1736 * set first. */
1737 err = -EINVAL;
1738 }
32a7627c 1739 if (err)
3178b0db 1740 goto error;
32a7627c 1741
624ce4f5 1742 bitmap->daemon_lastrun = jiffies;
42a04b50 1743 bitmap->chunkshift = ffz(~mddev->bitmap_info.chunksize);
32a7627c
N
1744
1745 /* now that chunksize and chunkshift are set, we can use these macros */
ac2f40be 1746 chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) >>
1f593903 1747 CHUNK_BLOCK_SHIFT(bitmap);
ac2f40be 1748 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
32a7627c
N
1749
1750 BUG_ON(!pages);
1751
1752 bitmap->chunks = chunks;
1753 bitmap->pages = pages;
1754 bitmap->missing_pages = pages;
1755 bitmap->counter_bits = COUNTER_BITS;
1756
1757 bitmap->syncchunk = ~0UL;
1758
44456d37 1759#ifdef INJECT_FATAL_FAULT_1
32a7627c
N
1760 bitmap->bp = NULL;
1761#else
9ffae0cf 1762 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
32a7627c 1763#endif
3178b0db 1764 err = -ENOMEM;
32a7627c 1765 if (!bitmap->bp)
3178b0db 1766 goto error;
32a7627c 1767
69e51b44
N
1768 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1769 pages, bmname(bitmap));
1770
1771 mddev->bitmap = bitmap;
1772
1773
1774 return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0;
1775
1776 error:
1777 bitmap_free(bitmap);
1778 return err;
1779}
1780
1781int bitmap_load(mddev_t *mddev)
1782{
1783 int err = 0;
1784 sector_t sector = 0;
1785 struct bitmap *bitmap = mddev->bitmap;
1786
1787 if (!bitmap)
1788 goto out;
1789
1790 /* Clear out old bitmap info first: Either there is none, or we
1791 * are resuming after someone else has possibly changed things,
1792 * so we should forget old cached info.
1793 * All chunks should be clean, but some might need_sync.
1794 */
1795 while (sector < mddev->resync_max_sectors) {
57dab0bd 1796 sector_t blocks;
69e51b44
N
1797 bitmap_start_sync(bitmap, sector, &blocks, 0);
1798 sector += blocks;
1799 }
1800 bitmap_close_sync(bitmap);
1801
e384e585
N
1802 if (mddev->bitmap_info.log) {
1803 unsigned long i;
1804 struct dm_dirty_log *log = mddev->bitmap_info.log;
1805 for (i = 0; i < bitmap->chunks; i++)
1806 if (!log->type->in_sync(log, i, 1))
1807 bitmap_set_memory_bits(bitmap,
1808 (sector_t)i << CHUNK_BLOCK_SHIFT(bitmap),
1809 1);
69e51b44
N
1810 } else {
1811 sector_t start = 0;
1812 if (mddev->degraded == 0
1813 || bitmap->events_cleared == mddev->events)
1814 /* no need to keep dirty bits to optimise a
1815 * re-add of a missing device */
1816 start = mddev->recovery_cp;
193f1c93 1817
69e51b44
N
1818 err = bitmap_init_from_disk(bitmap, start);
1819 }
32a7627c 1820 if (err)
69e51b44 1821 goto out;
3178b0db 1822
1b04be96 1823 mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
9cd30fdc 1824 md_wakeup_thread(mddev->thread);
b15c2e57 1825
4ad13663
N
1826 bitmap_update_sb(bitmap);
1827
69e51b44
N
1828 if (bitmap->flags & BITMAP_WRITE_ERROR)
1829 err = -EIO;
1830out:
3178b0db 1831 return err;
32a7627c 1832}
69e51b44 1833EXPORT_SYMBOL_GPL(bitmap_load);
32a7627c 1834
43a70507
N
1835static ssize_t
1836location_show(mddev_t *mddev, char *page)
1837{
1838 ssize_t len;
ac2f40be 1839 if (mddev->bitmap_info.file)
43a70507 1840 len = sprintf(page, "file");
ac2f40be 1841 else if (mddev->bitmap_info.offset)
43a70507 1842 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
ac2f40be 1843 else
43a70507
N
1844 len = sprintf(page, "none");
1845 len += sprintf(page+len, "\n");
1846 return len;
1847}
1848
1849static ssize_t
1850location_store(mddev_t *mddev, const char *buf, size_t len)
1851{
1852
1853 if (mddev->pers) {
1854 if (!mddev->pers->quiesce)
1855 return -EBUSY;
1856 if (mddev->recovery || mddev->sync_thread)
1857 return -EBUSY;
1858 }
1859
1860 if (mddev->bitmap || mddev->bitmap_info.file ||
1861 mddev->bitmap_info.offset) {
1862 /* bitmap already configured. Only option is to clear it */
1863 if (strncmp(buf, "none", 4) != 0)
1864 return -EBUSY;
1865 if (mddev->pers) {
1866 mddev->pers->quiesce(mddev, 1);
1867 bitmap_destroy(mddev);
1868 mddev->pers->quiesce(mddev, 0);
1869 }
1870 mddev->bitmap_info.offset = 0;
1871 if (mddev->bitmap_info.file) {
1872 struct file *f = mddev->bitmap_info.file;
1873 mddev->bitmap_info.file = NULL;
1874 restore_bitmap_write_access(f);
1875 fput(f);
1876 }
1877 } else {
1878 /* No bitmap, OK to set a location */
1879 long long offset;
1880 if (strncmp(buf, "none", 4) == 0)
1881 /* nothing to be done */;
1882 else if (strncmp(buf, "file:", 5) == 0) {
1883 /* Not supported yet */
1884 return -EINVAL;
1885 } else {
1886 int rv;
1887 if (buf[0] == '+')
1888 rv = strict_strtoll(buf+1, 10, &offset);
1889 else
1890 rv = strict_strtoll(buf, 10, &offset);
1891 if (rv)
1892 return rv;
1893 if (offset == 0)
1894 return -EINVAL;
ece5cff0
N
1895 if (mddev->bitmap_info.external == 0 &&
1896 mddev->major_version == 0 &&
43a70507
N
1897 offset != mddev->bitmap_info.default_offset)
1898 return -EINVAL;
1899 mddev->bitmap_info.offset = offset;
1900 if (mddev->pers) {
1901 mddev->pers->quiesce(mddev, 1);
1902 rv = bitmap_create(mddev);
1903 if (rv) {
1904 bitmap_destroy(mddev);
1905 mddev->bitmap_info.offset = 0;
1906 }
1907 mddev->pers->quiesce(mddev, 0);
1908 if (rv)
1909 return rv;
1910 }
1911 }
1912 }
1913 if (!mddev->external) {
1914 /* Ensure new bitmap info is stored in
1915 * metadata promptly.
1916 */
1917 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1918 md_wakeup_thread(mddev->thread);
1919 }
1920 return len;
1921}
1922
1923static struct md_sysfs_entry bitmap_location =
1924__ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
1925
1926static ssize_t
1927timeout_show(mddev_t *mddev, char *page)
1928{
1929 ssize_t len;
1930 unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
1931 unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
ac2f40be 1932
43a70507
N
1933 len = sprintf(page, "%lu", secs);
1934 if (jifs)
1935 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
1936 len += sprintf(page+len, "\n");
1937 return len;
1938}
1939
1940static ssize_t
1941timeout_store(mddev_t *mddev, const char *buf, size_t len)
1942{
1943 /* timeout can be set at any time */
1944 unsigned long timeout;
1945 int rv = strict_strtoul_scaled(buf, &timeout, 4);
1946 if (rv)
1947 return rv;
1948
1949 /* just to make sure we don't overflow... */
1950 if (timeout >= LONG_MAX / HZ)
1951 return -EINVAL;
1952
1953 timeout = timeout * HZ / 10000;
1954
1955 if (timeout >= MAX_SCHEDULE_TIMEOUT)
1956 timeout = MAX_SCHEDULE_TIMEOUT-1;
1957 if (timeout < 1)
1958 timeout = 1;
1959 mddev->bitmap_info.daemon_sleep = timeout;
1960 if (mddev->thread) {
1961 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
1962 * the bitmap is all clean and we don't need to
1963 * adjust the timeout right now
1964 */
1965 if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
1966 mddev->thread->timeout = timeout;
1967 md_wakeup_thread(mddev->thread);
1968 }
1969 }
1970 return len;
1971}
1972
1973static struct md_sysfs_entry bitmap_timeout =
1974__ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
1975
1976static ssize_t
1977backlog_show(mddev_t *mddev, char *page)
1978{
1979 return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
1980}
1981
1982static ssize_t
1983backlog_store(mddev_t *mddev, const char *buf, size_t len)
1984{
1985 unsigned long backlog;
1986 int rv = strict_strtoul(buf, 10, &backlog);
1987 if (rv)
1988 return rv;
1989 if (backlog > COUNTER_MAX)
1990 return -EINVAL;
1991 mddev->bitmap_info.max_write_behind = backlog;
1992 return len;
1993}
1994
1995static struct md_sysfs_entry bitmap_backlog =
1996__ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
1997
1998static ssize_t
1999chunksize_show(mddev_t *mddev, char *page)
2000{
2001 return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2002}
2003
2004static ssize_t
2005chunksize_store(mddev_t *mddev, const char *buf, size_t len)
2006{
2007 /* Can only be changed when no bitmap is active */
2008 int rv;
2009 unsigned long csize;
2010 if (mddev->bitmap)
2011 return -EBUSY;
2012 rv = strict_strtoul(buf, 10, &csize);
2013 if (rv)
2014 return rv;
2015 if (csize < 512 ||
2016 !is_power_of_2(csize))
2017 return -EINVAL;
2018 mddev->bitmap_info.chunksize = csize;
2019 return len;
2020}
2021
2022static struct md_sysfs_entry bitmap_chunksize =
2023__ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2024
ece5cff0
N
2025static ssize_t metadata_show(mddev_t *mddev, char *page)
2026{
2027 return sprintf(page, "%s\n", (mddev->bitmap_info.external
2028 ? "external" : "internal"));
2029}
2030
2031static ssize_t metadata_store(mddev_t *mddev, const char *buf, size_t len)
2032{
2033 if (mddev->bitmap ||
2034 mddev->bitmap_info.file ||
2035 mddev->bitmap_info.offset)
2036 return -EBUSY;
2037 if (strncmp(buf, "external", 8) == 0)
2038 mddev->bitmap_info.external = 1;
2039 else if (strncmp(buf, "internal", 8) == 0)
2040 mddev->bitmap_info.external = 0;
2041 else
2042 return -EINVAL;
2043 return len;
2044}
2045
2046static struct md_sysfs_entry bitmap_metadata =
2047__ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2048
2049static ssize_t can_clear_show(mddev_t *mddev, char *page)
2050{
2051 int len;
2052 if (mddev->bitmap)
2053 len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2054 "false" : "true"));
2055 else
2056 len = sprintf(page, "\n");
2057 return len;
2058}
2059
2060static ssize_t can_clear_store(mddev_t *mddev, const char *buf, size_t len)
2061{
2062 if (mddev->bitmap == NULL)
2063 return -ENOENT;
2064 if (strncmp(buf, "false", 5) == 0)
2065 mddev->bitmap->need_sync = 1;
2066 else if (strncmp(buf, "true", 4) == 0) {
2067 if (mddev->degraded)
2068 return -EBUSY;
2069 mddev->bitmap->need_sync = 0;
2070 } else
2071 return -EINVAL;
2072 return len;
2073}
2074
2075static struct md_sysfs_entry bitmap_can_clear =
2076__ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2077
696fcd53
PC
2078static ssize_t
2079behind_writes_used_show(mddev_t *mddev, char *page)
2080{
2081 if (mddev->bitmap == NULL)
2082 return sprintf(page, "0\n");
2083 return sprintf(page, "%lu\n",
2084 mddev->bitmap->behind_writes_used);
2085}
2086
2087static ssize_t
2088behind_writes_used_reset(mddev_t *mddev, const char *buf, size_t len)
2089{
2090 if (mddev->bitmap)
2091 mddev->bitmap->behind_writes_used = 0;
2092 return len;
2093}
2094
2095static struct md_sysfs_entry max_backlog_used =
2096__ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2097 behind_writes_used_show, behind_writes_used_reset);
2098
43a70507
N
2099static struct attribute *md_bitmap_attrs[] = {
2100 &bitmap_location.attr,
2101 &bitmap_timeout.attr,
2102 &bitmap_backlog.attr,
2103 &bitmap_chunksize.attr,
ece5cff0
N
2104 &bitmap_metadata.attr,
2105 &bitmap_can_clear.attr,
696fcd53 2106 &max_backlog_used.attr,
43a70507
N
2107 NULL
2108};
2109struct attribute_group md_bitmap_group = {
2110 .name = "bitmap",
2111 .attrs = md_bitmap_attrs,
2112};
2113
This page took 0.670496 seconds and 5 git commands to generate.